OpenAI, Anthropic restrict new AI models amid Trump cyber review

OpenAI held back its new GPT-5.6 Sol model for a small group of trusted partners at the request of the Trump administration, while Anthropic won permission to restore limited access to its strongest cybersecurity model, Mythos 5, for more than 100 U.S. companies and government agencies. The twin moves marked a rare moment when access to cutting-edge commercial AI was filtered through government approval, with Washington forcing a staged rollout rather than allowing both companies to release their latest models broadly.

OpenAI said it had previewed GPT-5.6 Sol’s capabilities and shared its launch plans with the U.S. government before release. The company plans to make the model broadly available in the coming weeks, but said it does not want government-vetting restrictions to become the long-term default. The limited rollout covered GPT-5.6 Sol and two other models in the same lineup, Terra and Luna, after the White House asked for restraint because of the model’s advanced capabilities and cybersecurity concerns. OpenAI had not initially intended to restrict the general-use model, but changed course after consultations with the Office of Science and Technology Policy and the Office of the National Cyber Director.

Anthropic moved almost immediately after OpenAI’s announcement, saying the Trump administration had approved a limited re-release of Mythos 5 after the Commerce Department effectively blocked it about two weeks earlier. The company said the model can again be used by trusted partners and cyber defenders, but its less powerful Fable 5 model remains blocked. Anthropic said the U.S. government had cited national security authorities when it issued an export-control directive on June 12, 2026, suspending all access to Fable 5 and Mythos 5 by any foreign national, including foreign national employees inside the company and abroad.

Anthropic said it had been working with the government since June 12 to restore access. The company’s limited reinstatement, paired with OpenAI’s restricted release, underscored how the administration has moved from broad AI safety rhetoric to direct gatekeeping of commercial technology. For companies building frontier models in Washington’s orbit, the immediate question is no longer only how powerful the systems are, but who gets to use them, when, and under what federal authority.